pcp
[Top] [All Lists]

pcp updates: brolley+emery+mgoodwin merges, qa

To: PCP Mailing List <pcp@xxxxxxxxxxx>
Subject: pcp updates: brolley+emery+mgoodwin merges, qa
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Mon, 14 Jul 2014 23:49:59 -0400 (EDT)
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <800683057.10166653.1405396095376.JavaMail.zimbra@xxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: LxkhdyZtJy8PcbfBd1hqZazXeyjl+Q==
Thread-topic: pcp updates: brolley+emery+mgoodwin merges, qa
Changes committed to git://git.performancecopilot.org/pcp.git dev

 Makepkgs                          |    9 ++++--
 qa/061                            |    6 +++-
 qa/061.out                        |   50 +++++++++++++++++++-------------------
 src/libpcp/src/auxconnect.c       |    8 +++++-
 src/libpcp/src/internal.h         |    1 
 src/libpcp/src/secureconnect.c    |   28 ++++++++++++---------
 src/libpcp_gui/src/timeclient.c   |   24 +-----------------
 src/perl/PMDA/local.c             |   42 -------------------------------
 src/pmcd/src/config.c             |   21 ---------------
 src/pmdas/lustrecomm/lustrecomm.c |   10 +++----
 src/pmdas/sample/src/sample.c     |    2 -
 11 files changed, 69 insertions(+), 132 deletions(-)

commit 07423d8eaf0f9667001f1934bf6ae916dd759fad
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Tue Jul 15 13:47:09 2014 +1000

    pmdasample: trivial strcpy tweak to defend against older valgrind
    
    On a certain far-flung 64bit architecture, on a relatively old
    enterprise distro, valgrind is producing a bogus warning for an
    eight-byte read on a 3byte strcpy.  Use strncpy to dodge this.

commit cdfa900f4286f4eaf79bfa819d488180f2fa4a4d
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Tue Jul 15 13:44:06 2014 +1000

    Fix qa/061 filter - long hostnames causing pmresult differences
    
    Test 061 can produce different output on different hosts as a result
    of the log preamble result write (v2 logs with pmcd.pmlogger.* in an
    initial result).  Filter the first result write size but keep all of
    the subsequent result sizes in the output, as these are under test.

commit 456432a4a2154adf03c6d57752420eacece26b52
Author: Mark Goodwin <mgoodwin@xxxxxxxxxx>
Date:   Tue Jul 15 10:56:15 2014 +1000

    Fix Makepkgs build on platforms with old GNU tar.
    
    Older versions of tar (e.g. on RHEL5), do not support --transform
    so we have to use git-archive to construct the src tarball. Builds
    on affected platforms will thus not include any local changes.
    
        modified:   Makepkgs

commit 955cb99b2549066d239e77105f5e6304c4a4e55f
Merge: f11bfb6 122effc
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Tue Jul 15 10:22:08 2014 +1000

    Merge branch 'brolley/dev' of ../pcpfans into brolley-merge

commit f11bfb661878b96cd081b1eeef173e3a797b29f5
Merge: 9279ce5 c02cc9a
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Tue Jul 15 08:57:07 2014 +1000

    Merge branch 'dev' of git://git.performancecopilot.org/emery/pcp into 
emery-merge

commit 122effcc48ba680e67311156e6ee61ae28922df1
Author: Dave Brolley <brolley@xxxxxxxxxx>
Date:   Mon Jul 14 13:57:39 2014 -0400

    Abstract __pmConnect() into __pmConnect and __pmConnectWithFNDELAY().
    
    In PCP, control over whether a connection attempt made via
    __pmConnect() blocks or not is controlled by setting the
    FNDELAY (O_NONBLOCK) file status flag in for the file
    descriptor. However, NSPR's PR_Connect ignores the FNDELAY
    flags (more accurately, always sets it) and uses an explicit
    timeout argument to control whether the connection attempt
    blocks and, if so, for how long. The NSPR implementation of
    __pmConnect(), therefore has no way of knowing what the caller's
    intent was.
    
    To solve this, we need an explicit __pmConnectWithFNDELAY(). This
    new function is internal to libpcp and not exported. It is called
    from within __pmConnectTo() which is the function called within
    PCP when non-blocking behavior is needed.
    
    Code within PCP that calls __pmConnect() directly does not expect
    non-blocking behavior and so does not need to handle EINPROGRESS.
    This commit reverts a previous change
    (commit commit a705d6bb86d87fa0cdf0fb0accf5299212cef08d)
    for which such handling was added.
    
    This commit fixes regressions in qa tests 230 and 835.

commit c02cc9a1ea5f5e3697644b6f27811bf44315b0b2
Author: Scott Emery <emery@xxxxxxx>
Date:   Fri Jul 11 16:05:59 2014 -0500

    src/pmdas/lustrecomm/lustrecomm.c
    lustrecomm would abort if any lustrecomm.stats.* were collected when
    one or more of lustrecom.stats.*_length exceeded 32 bits. Defined as
    PM_TYPE_64, called file_indexed with PM_TYPE_32, which used
    strtoul on 64 bit value.

commit e07d8e8e17755578f7f5f9929b6506ca798c1798
Author: Scott Emery <emery@xxxxxxx>
Date:   Fri Jul 11 16:01:15 2014 -0500

    pcp/src/pmdas/lustrecomm.c
    lustrecomm.lnet_memused cannot open file.
    file name was incorrect.

<Prev in Thread] Current Thread [Next in Thread>
  • pcp updates: brolley+emery+mgoodwin merges, qa, Nathan Scott <=